WO2021253671A1

WO2021253671A1 - Magnetic resonance cine imaging method and apparatus, and imaging device and storage medium

Info

Publication number: WO2021253671A1
Application number: PCT/CN2020/117462
Authority: WO
Inventors: 梁栋; 朱燕杰; 柯子文; 刘新; 郑海荣
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2020-06-18
Filing date: 2020-09-24
Publication date: 2021-12-23
Also published as: CN111714124A; CN111714124B

Abstract

A magnetic resonance cine imaging method and apparatus, and an imaging device and a storage medium. The method comprises: acquiring magnetic resonance data (S101); and inputting the magnetic resonance data into a trained imaging model to obtain a magnetic resonance cine image, wherein the imaging model is a sparse low-rank network model constructed on the basis of an alternating direction method of multipliers and is used for controlling the alternating direction method of multipliers to perform iterative solving according to iteration parameters that are output by a neural network model, so as to obtain the magnetic resonance cine image (S102).

Description

Magnetic resonance film imaging method, device, imaging equipment and storage medium

This disclosure claims the priority of a Chinese patent application filed with the Chinese Patent Office with an application number of 202010560886.0 on June 18, 2020, and the entire content of the above application is incorporated into this disclosure by reference.

Technical field

This application relates to the field of magnetic resonance imaging, for example, to a magnetic resonance film imaging method, device, imaging equipment, and storage medium.

Background technique

Magnetic resonance cardiac cine imaging is a non-invasive imaging technique that can be used to assess cardiac function, abnormal ventricular wall motion, etc., and provide a wealth of information for clinical diagnosis of the heart. However, due to the constraints of magnetic resonance physics, hardware, and the length of the cardiac motion cycle, magnetic resonance cardiac cine imaging is often limited in terms of time and space resolution, and cannot accurately assess some heart diseases, such as arrhythmia. Therefore, under the premise of ensuring the imaging quality, it is particularly important to improve the speed and spatial resolution of MRI cardiac cine imaging.

In recent years, many people have been exploring the use of deep learning methods in the field of magnetic resonance film imaging. For example, magnetic resonance film imaging based on cascaded convolutional network (DC-CNN), convolutional recurrent neural network (CRNN), and multi-supervised cross-domain network DIMENSION have achieved good reconstruction results. However, because these three kinds of neural networks directly learn the mapping relationship from under-collected images to full-collected images, they require a longer reconstruction time during the image reconstruction process, or the quality of the reconstructed MRI cardiac film images is low.

In summary, the deep learning methods of related technologies cannot simultaneously take into account the image reconstruction time and image quality in the field of magnetic resonance film imaging.

Summary of the invention

The embodiments of the present application provide a magnetic resonance film imaging method, device, imaging device, and storage medium, which solve the problem that deep learning of related technologies cannot simultaneously take into account the image reconstruction time and image quality in the field of magnetic resonance film imaging.

In the first aspect, an embodiment of the present application provides a magnetic resonance cine imaging method, including:

Obtain magnetic resonance data;

The magnetic resonance data is input into a trained imaging model to obtain a magnetic resonance movie image, wherein the imaging model is a sparse low-rank network model constructed based on the alternating direction multiplier algorithm, and the imaging model is used to control the alternating The direction multiplier algorithm is iteratively solved according to the iterative parameters output by the neural network model to obtain the magnetic resonance movie image.

In the second aspect, an embodiment of the present application also provides a magnetic resonance film imaging device, including:

The obtaining module is configured to obtain magnetic resonance data;

The reconstruction module is configured to control the neural network model to determine the iteration parameters required for the current iteration solution according to the previous iteration solution result of the alternate direction multiplier algorithm, and to control the alternate direction multiplier algorithm to complete the current iteration solution process according to the iteration parameters, Until the current iterative solution result meets the preset convergence condition.

In a third aspect, an embodiment of the present application also provides an imaging device, the imaging device including:

processor;

Storage device for storing programs;

When the program is executed by the processor, the processor realizes the magnetic resonance cine imaging method according to any embodiment.

In a fourth aspect, an embodiment of the present application also provides a storage medium containing computer-executable instructions, which are used to execute the magnetic resonance cine imaging method as described in any of the embodiments when the computer-executable instructions are executed by a computer processor .

Description of the drawings

The following will briefly introduce the drawings that need to be used in the description of the embodiments. The drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

FIG. 1 is a flowchart of a magnetic resonance cine imaging method provided in Embodiment 1 of the present application;

2 is a schematic diagram of iterative ADMM algorithm provided by Embodiment 1 of the present application;

FIG. 3A is a structural block diagram of a magnetic resonance cine imaging apparatus provided by Embodiment 2 of the present application;

3B is a structural block diagram of another magnetic resonance cine imaging apparatus provided by the second embodiment of the present application;

FIG. 4 is a structural block diagram of an imaging device provided in Embodiment 3 of the present application.

detailed description

The following will clearly and completely describe the technical solutions of the present application through implementations with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are a part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by a person of ordinary skill in the art without creative work shall fall within the protection scope of this application.

Example one

FIG. 1 is a flowchart of a magnetic resonance cine imaging method provided in Embodiment 1 of the present application. The technical solution of this embodiment is applicable to the case where a sparse low-rank network model constructed based on the ADMM (Alternating Direction Method of Multipliers) algorithm is used to quickly reconstruct a high-quality magnetic resonance movie image. The method may be executed by the magnetic resonance cine imaging apparatus provided in the embodiment of the present application, and the apparatus may be implemented in a software and/or hardware manner, and configured to be applied in a processor of an imaging device. The method may include the following steps:

S101. Acquire magnetic resonance data.

The magnetic resonance data in this embodiment is a magnetic resonance signal containing time information. For example, cardiac magnetic resonance signals.

S102. Input the magnetic resonance data into the trained imaging model to obtain a magnetic resonance movie image, where the imaging model is a sparse low-rank network model constructed based on the ADMM algorithm, and is used to control the iterative parameters output by the ADMM algorithm according to the neural network model Carry out the corresponding iterative solution to obtain the magnetic resonance film image.

After the magnetic resonance data is obtained, it is input into a trained imaging model, and the trained imaging model analyzes it to obtain a magnetic resonance movie image.

In one embodiment, the imaging model is a sparse low-rank network model constructed based on the ADMM algorithm, and may be a model that combines the iterative solution process of the ADMM algorithm and the neural network model. In the imaging model, the neural network model is set to determine the iterative parameters required for the current iterative solution of the ADMM algorithm according to the previous iterative solution result of the ADMM algorithm, until the current iterative solution result meets the preset convergence condition. If the current iterative solution result meets the preset convergence condition, the neural network model ends the calculation of the iterative parameters, and the iterative solution result is the magnetic resonance film image to be output by the imaging model.

In one embodiment, the method for constructing the imaging model includes: converting the under-collected reconstruction task of the magnetic resonance signal to iteratively solving the data consistency sub-problem, the low-rank sub-problem, the sparse sub-problem, and the auxiliary variable sub-problem based on the ADMM algorithm; and the control neural network The model determines the iterative parameters required for the current iterative solution according to the previous iterative solution results of the data consistency sub-problem, low-rank sub-problem, sparse sub-problem, and auxiliary variable sub-problem, and controls the ADMM algorithm to complete the current iterative solution process according to the iterative parameters , Until the current iterative solution result meets the preset convergence condition.

In some embodiments, based on the ADMM algorithm, the under-collection reconstruction task of the magnetic resonance signal is converted into the step of solving the data consistency sub-problem, the low-rank sub-problem, the sparse sub-problem and the auxiliary variable sub-problem, including: the under-collection of the magnetic resonance signal The reconstruction task is modeled as the data consistency constraint problem, the transform domain sparse constraint problem and the low rank constraint problem; then the data consistency problem, the transform domain sparse problem and the low rank constraint problem are solved based on the ADMM algorithm to reconstruct the under-collection of the magnetic resonance signal. It is transformed into solving data consistency sub-problems, low-rank sub-problems, sparse sub-problems and auxiliary variable sub-problems. The process can be as follows:

For magnetic resonance data, that is, K-space data

The corresponding under-picked heart movie image is

The under-taken reconstruction problem of the cardiac movie image can be modeled as the following optimization problem:

Among them, A=PF is the measurement matrix, P is the sampling matrix, F is the Fourier transform, D is the sparse transform, g(·) is the sparse constraint, ||·|| _* =∑ _i (Σ _i ,i) is The kernel function, Σ is a vector of singular values, the kernel norm is the sum of the first i largest singular values of the signal, reflecting the low-rank characteristics of the signal; λ ₁ and λ ₂ are both regularization coefficients.

Introducing the auxiliary variable z=Dx, t=x, the above optimization problem becomes:

The augmented Lagrangian form of formula (2) is as follows:

Among them, α ₁ is the Lagrangian multiplier in the sparse transformation, α ₂ is the Lagrangian multiplier in the sparse transformation, ρ ₁ is the penalty coefficient in the sparse transformation, and ρ ₂ is the penalty coefficient in the low-rank transformation. .

Using ADMM algorithm to solve formula (3), we can get:

Use transform

A=FP, t ⁽ⁿ⁺¹⁾ is represented by a singular value threshold, and the following four sub-problems are obtained.

Among them, IST means that the signal is decomposed by SVD (Singular Value Decomposition), that is, singular value decomposition, to obtain the eigenvalue vector, threshold the eigenvalue vector, and then restore to the original signal; S is a non-linear threshold function, used to The sparse matrix performs threshold filtering; λ1 is the regularization coefficient for sparse transformation, λ2 is the regularization coefficient for low-rank transformation, η ₁ is the update step in sparse transformation, and η ₂ is the update step in low-rank transformation. Long, I is the unit matrix of all 1, m and n are the size of x, σ is the singular value of x, u is the left singular value vector of x, v is the right singular value vector of x, and p is for The constant that controls the scale.

It is understandable that " ^T " represents the transposed matrix, for example, "P ^T "represents the transposed matrix of "P";" ^* " represents the adjoint matrix, for example, "v _i ^* " represents the adjoint matrix of _{"v i"} .

Refer to Figure 2 for the iterative solution steps of formula (5), and combine it with the neural network model. After the combination, the neural network model can use the previous iterative solution result of the ADMM algorithm to determine the iterative parameters required for its current iterative solution until the current iterative solution result meets the preset convergence conditions. It is understandable that the combination of the ADMM algorithm and the neural network model enables the neural network model to quickly and accurately determine the current iterative solution based on the prior knowledge of the learned magnetic resonance data in terms of sparseness and low rank Iteration parameters. Among them, the iteration parameters include regularization coefficients λ1, λ2 and sparse transformation D.

It is understandable that after the imaging model is constructed, it cannot be used directly for image reconstruction, and a certain number of samples need to be trained to generate a trained imaging model. After the trained imaging model is obtained, the trained imaging model can be used to reconstruct the magnetic resonance data to generate a magnetic resonance movie image.

The technical solutions of the magnetic resonance cine imaging method provided by the embodiments of the present application are compared with related technologies. Combining the ADMM algorithm with the neural network model allows the neural network model to learn the prior knowledge of sparse and low-rank magnetic resonance data, and use the learned prior knowledge to quickly and accurately determine the solution of the ADMM algorithm in each iteration The required iterative parameters until the iterative solution result of the ADMM algorithm meets the preset convergence conditions. Because the iterative parameter is determined more quickly and accurately, the time required for the image reconstruction process is greatly reduced, and the quality of the reconstructed magnetic resonance film image is greatly improved.

Example two

Fig. 3A is a structural block diagram of a magnetic resonance cine imaging apparatus provided by an embodiment of the present application. The device is used to execute the magnetic resonance cine imaging method provided in any of the foregoing embodiments, and the device can be implemented in software or hardware. The device includes:

The obtaining module 11 is configured to obtain magnetic resonance data;

The reconstruction module 12 is configured to input the magnetic resonance data into a trained imaging model to obtain a magnetic resonance movie image, wherein the imaging model is a sparse low-rank network model constructed based on the ADMM algorithm, and is used to control the ADMM The algorithm performs an iterative solution based on the iterative parameters output by the neural network model to obtain a magnetic resonance movie image.

Optionally, the device further includes a model building module 101 (see FIG. 3B), and the model building module includes:

The task conversion unit is configured to convert the under-collection reconstruction task of the magnetic resonance signal based on the ADMM algorithm to iteratively solve the data consistency sub-problem, the low-rank sub-problem, the sparse sub-problem and the auxiliary variable sub-problem;

The combination unit is configured to control the neural network model to determine the iteration parameters required for the current iteration solution according to the previous iteration solution result of the ADMM algorithm, and to control the ADMM algorithm to complete the current iteration solution process according to the iteration parameter until the current iteration solution result meets Preset convergence conditions.

Wherein, the task conversion unit can optionally be configured to model the under-collection reconstruction task of the magnetic resonance signal as a data consistency constraint problem, a transform domain sparse constraint problem, and a low-rank constraint problem; solve the data consistency problem, transform domain sparseness problem based on ADMM Problem and low-rank constraint problem to transform the under-collection reconstruction task of magnetic resonance signals into solving data consistency sub-problems, low-rank sub-problems, sparse sub-problems and auxiliary variable sub-problems.

Optionally, the device further includes a training module 102 (see FIG. 3B), which is configured to receive training sample data and complete the training of the imaging model according to the received training sample data to generate a trained imaging model.

The technical solutions of the magnetic resonance cine imaging apparatus provided by the embodiments of the present application are compared with related technologies. Combining the ADMM algorithm with the neural network model allows the neural network model to learn the prior knowledge of sparse and low-rank magnetic resonance data, and use the learned prior knowledge to quickly and accurately determine the solution of the ADMM algorithm in each iteration The required iterative parameters until the iterative solution result of the ADMM algorithm meets the preset convergence conditions. Because the iterative parameter is determined more quickly and accurately, the time required for the image reconstruction process is greatly reduced, and the quality of the reconstructed magnetic resonance film image is greatly improved.

The magnetic resonance cine imaging apparatus provided by the embodiment of the present application can execute the magnetic resonance cine imaging method provided by any embodiment of the present application, and has the corresponding functional modules and beneficial effects for the execution method.

Example three

FIG. 4 is a schematic structural diagram of an imaging device provided in Embodiment 3 of the application. As shown in FIG. 4, the device includes a processor 201, a memory 202, an input device 203, and an output device 204; the number of processors 201 in the device may be one Or more, one processor 201 is taken as an example in FIG. 4; the processor 201, the memory 202, the input device 203, and the output device 204 in the device may be connected by a bus or other methods. In FIG. 4, a bus connection is taken as an example.

As a computer-readable storage medium, the memory 202 can be used to store software programs, computer-executable programs, and modules, such as the magnetic resonance film imaging method, device, imaging device, and program instructions/modules corresponding to the storage medium in the embodiments of the present application (For example, acquisition module 11 and reconstruction module 12). The processor 201 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 202, that is, realizes the aforementioned magnetic resonance film imaging method, device, imaging device, and storage medium.

The memory 202 may mainly include a program storage area and a data storage area. The program storage area may store an operating system and an application program required by at least one function; the data storage area may store data created according to the use of the terminal, and the like. In addition, the memory 202 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some examples, the memory 202 may include a memory remotely provided with respect to the processor 201, and these remote memories may be connected to the device through a network. Examples of the aforementioned network may include the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The input device 203 can be used to receive inputted digital or character information, and generate key signal input related to user settings and function control of the device.

The output device 204 may include a display device such as a display screen, for example, a display screen of a user terminal.

Embodiment four

The fourth embodiment of the present application also provides a storage medium containing computer-executable instructions, when the computer-executable instructions are executed by a computer processor, they are used to execute a magnetic resonance film imaging method, device, imaging device, and storage medium , The method includes:

Obtain magnetic resonance data;

The control neural network model determines the iterative parameters required for the current iterative solution according to the previous iterative solution result of the ADMM algorithm, and controls the ADMM algorithm to complete the current iterative solution process according to the iterative parameter until the current iterative solution result meets the preset convergence condition.

Of course, a storage medium provided by an embodiment of the present application contains computer-executable instructions. The computer-executable instructions can execute the method operations described above, and can also execute the magnetic resonance cine imaging method provided by any embodiment of the present application. Related operations in.

Through the above description of the implementation manners, those skilled in the art can clearly understand that this application can be implemented by means of software and necessary general-purpose hardware, of course, it can also be implemented by hardware, but in many cases the former is a better implementation. . Based on this understanding, the technical solution of this application essentially or the part that contributes to the related technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, Read-Only Memory (ROM), Random Access Memory (RAM), Flash memory (FLASH), hard disk or optical disk, etc., including several instructions to make a computer device (which can be a personal A computer, a server, or a network device, etc.) execute the magnetic resonance film imaging method described in each embodiment of the present application.

It is worth noting that, in the above-mentioned embodiment of the magnetic resonance film imaging device, the various units and modules included are only divided according to the functional logic, as long as the corresponding functions can be realized; in addition, the names of the functional units are only for It is easy to distinguish each other.

The technical solution of the magnetic resonance film imaging method provided by the embodiments of the present application includes: acquiring magnetic resonance data; inputting the magnetic resonance data into a trained imaging model to obtain a magnetic resonance film image, wherein the imaging model is constructed based on the ADMM algorithm The sparse low-rank network model of the imaging model is used to control the ADMM algorithm to perform corresponding iterative solutions according to the iterative parameters output by the neural network model to obtain the magnetic resonance movie image. Compared to related technologies. Combining the ADMM algorithm with the neural network model allows the neural network model to learn the prior knowledge of sparse and low-rank magnetic resonance data, and use the learned prior knowledge to quickly and accurately determine the solution of the ADMM algorithm in each iteration The required iterative parameters until the iterative solution result of the ADMM algorithm meets the preset convergence conditions. Because the iterative parameter is determined more quickly and accurately, the time required for the image reconstruction process is greatly reduced, and the quality of the reconstructed magnetic resonance film image is greatly improved.

Claims

A magnetic resonance film imaging method, including:

Obtain magnetic resonance data;

The magnetic resonance data is input into a trained imaging model to obtain a magnetic resonance movie image, wherein the imaging model is a sparse low-rank network model constructed based on the alternating direction multiplier algorithm, and the imaging model is used to control the alternating The direction multiplier algorithm is iteratively solved according to the iterative parameters output by the neural network model to obtain the magnetic resonance movie image.
The magnetic resonance film imaging method according to claim 1, wherein the neural network model is used to determine the iteration parameters required for the current iterative solution of the alternating direction multiplier algorithm according to the previous iteration solution result of the alternating direction multiplier algorithm.
The magnetic resonance cine imaging method according to claim 1, wherein the method for constructing the imaging model comprises:

Based on the alternating direction multiplier algorithm, the task of under-collected reconstruction of magnetic resonance signals is transformed into iterative solution of data consistency sub-problems, low-rank sub-problems, sparse sub-problems and auxiliary variable sub-problems;

The control neural network model determines the iteration parameters required for the current iteration solution according to the previous iteration solution results of the data consistency sub-problem, low-rank sub-problem, sparse sub-problem, and auxiliary variable sub-problem, and controls the alternating direction multiplier algorithm according to the iteration The parameter completes the current iterative solution process until the current iterative solution result meets the preset convergence condition.
The magnetic resonance film imaging method according to claim 3, wherein the algorithm based on the alternating direction multiplier transforms the under-taken reconstruction task of the magnetic resonance signal into solving the data consistency sub-problem, the low-rank sub-problem, the sparse sub-problem and the auxiliary variable Sub-questions, including:

The under-collection reconstruction task of magnetic resonance signals is modeled as a data consistency constraint problem, a transform domain sparse constraint problem, and a low-rank constraint problem;

Solving the data consistency problem, transform domain sparseness problem and low rank constraint problem based on alternating direction multipliers, so as to transform the under-collection reconstruction task of magnetic resonance signals into solving data consistency sub-problems, low-rank sub-problems, sparse sub-problems and auxiliary variables Sub-problem.
The magnetic resonance cine imaging method according to claim 3, wherein:

The data consistency sub-problems are:

The low-rank sub-problem is:

The sparse sub-problem is:

The auxiliary variable update sub-problem is:

Among them, y is the magnetic resonance signal, x is the image data, P is the sampling matrix, F is the Fourier transform; z and t are auxiliary variables, z=Dx, t=x, D is the sparse transform;
α 1 is the Lagrangian multiplier in the sparse transformation, α 2 is the Lagrangian multiplier in the low-rank transformation; ρ 1 is the penalty coefficient in the sparse transformation, and ρ 2 is the penalty coefficient in the low-rank transformation; IST means that the singular value decomposition of the signal is performed to obtain the eigenvalue vector, and the eigenvalue vector is thresholded, and then restored to the original signal; S is the nonlinear threshold function; λ1 is the regularization coefficient used for sparse transformation, and λ2 is used for Regularization coefficient of low-rank transformation, η 1 is the update step size in sparse transformation, η 2 is the update step size in low-rank transformation, I is the unit matrix of all 1, m and n are the size of x, and σ is x U is the left singular value vector of x, v is the right singular value vector of x, and p is a constant used to control the scale.
The magnetic resonance cine imaging method according to claim 5, wherein the iteration parameters include D, λ 1 and λ 2 .
The magnetic resonance film imaging method according to any one of claims 1 to 6, wherein the magnetic resonance data is cardiac magnetic resonance data.
A magnetic resonance film imaging device, including:

The obtaining module is configured to obtain magnetic resonance data;

The reconstruction module is configured to control the neural network model to determine the iteration parameters required for the current iteration solution according to the previous iteration solution result of the alternate direction multiplier algorithm, and to control the alternate direction multiplier algorithm to complete the current iteration solution process according to the iteration parameters, Until the current iterative solution result meets the preset convergence condition.
An imaging device, the imaging device comprising:

processor;

Storage device for storing programs;

When the program is executed by the processor, the processor realizes the magnetic resonance cine imaging method according to any one of claims 1-7.
A storage medium containing computer-executable instructions, when the computer-executable instructions are executed by a computer processor, are used to execute the magnetic resonance cine imaging method according to any one of claims 1-7.